Enhanced prediction performance of a time-delay reservoir computing system based on a VCSEL by dual-training method

Area of Science:

• Optoelectronics
• Nonlinear Dynamics
• Machine Learning

Background:

• Reservoir computing (RC) offers a powerful framework for processing complex time-series data.
• Traditional RC methods often face limitations in prediction accuracy for intricate systems.
• Vertical-cavity surface-emitting lasers (VCSELs) are promising optoelectronic components for implementing RC systems.

Purpose of the Study:

• To introduce and experimentally validate a new dual-training method for time-delay reservoir computing (RC) systems.
• To investigate the prediction performance enhancement achieved by the dual-training method.
• To analyze the impact of various system parameters on the performance of the proposed method.

Main Methods:

• Development of a dual-training algorithm for RC systems.
• Experimental implementation of the dual-training RC system using a single VCSEL.
• Numerical simulations to complement experimental findings.
• Systematic investigation of parameters including injection power, bias currents, feedback strength, and frequency detuning.

Main Results:

• The dual-training method significantly enhances the prediction performance of the VCSEL-based RC system.
• Experimental results show up to a 760% improvement in normalized mean squared error (NMSE) compared to single-training methods.
• The study identifies optimal parameter ranges for improved system performance.

Conclusions:

• The proposed dual-training method offers a substantial performance boost for RC systems.
• This advancement holds significant potential for future applications of RC in complex signal processing and prediction tasks.
• The experimental demonstration validates the efficacy of the dual-training approach in optoelectronic RC systems.